Acoustic echo cancellation with double-talk detection and acoustic channel variation

Abstract

In this thesis, the main goal is the adaptive identification of the long acoustic impulse responses, in the context of the acoustic echo cancellation (AEC), by fast and low complexity algorithms and in the case of a variable acoustic channel and double-talk scenario. Two major contributions were introduced; the first one is to propose two algorithms based on the set-membership identification (SMI), the purpose is to improve the convergence speed and the tracking capability with robustness against additive noise. The second contribution is the integration of double-talk detector (DTD) based on normalized cross-correlation (NCC) in the two proposed algorithms to tackle the problem of the the acoustic echo in the case of the presence of double-talk. The obtained results demonstrate better performances with lower computational complexity in the single-talk situation, also in the double-talk situation the proposed algorithms present good performances using DTD compared to the existing algorithms.